Q-learning-based hyper-heuristic algorithm for priority and precedence dual-driven task assignment in spatial crowdsourcing

Abstract

In spatial crowdsourcing, a core issue is to formulate an effective task assignment plan, based on the bipartite matching between the two parties, i.e., workers and tasks. In this context, one key challenge is how to suitably assign tasks to available workers and determine their execution order, with the consideration of the priority of all tasks, under the precedence constraint of tasks. To this end, we investigate an important problem, namely, priority and precedence dual-driven task assignment problem in spatial crowdsourcing (PDTAP-SC). A Q-learning-based hyper-heuristic (QLHH) algorithm is proposed to address this problem, which strives to simultaneously minimize the task completion time (i.e., makespan) and the overall completion time of all priority tasks. Specifically, QLHH utilizes a Q-learning-based high-level strategy to autonomously choose appropriate heuristics from a predefined set of low-level heuristics. At various stages of the optimization process, the chosen heuristic is treated as an executable action and applied to the solution space for better results. Moreover, critical configurations of parameters are systematically analyzed by conducting a design-of-experiment (DOE) approach. Finally, as a verification, both computational simulation and comparison are carried out in cases of different scales collected from a synthetic dataset, which is created by extending a real dataset, and the results demonstrate the effectiveness and efficiency of the proposed QLHH.